Introduction
ZunPowerTech, a company pioneers in on-demand delivery with solar vehicles, requested from us to develop an innovative Artificial Intelligence solution aimed at reducing operational costs. We analyzed various solutions and developed a tailor-made strategy to achieve the maximum solar energy capture during the journey of vehicles with solar panels. We had to process satellite data, and detect the best routes for the vehicles.
My role and contributions
- As a Principal Software Engineer, I directed the AI team, comprising 4 professionals and managed the requirements and functionalities needs with the business.
- I led the analysis, architecture, design, documentation, development, quality assurance of the new solution, and satellite cloud data parsing.
How we did it
The first step was to analyze all possible solutions to reduce operation costs, and after understanding the customer business flow and many attempts analyzing different solutions, we found out a generic solution that covered the main requirement. The project development had the following phases in order to develop an MVP product which was able to detect the best possible routing for vehicles depending on the solar energy:
Data gathering
- Getting the data maps of the city of Madrid, with all their buildings in OpenStreetMaps.
- Getting the OSIF cloud satellite data of the buildings of Madrid, in order to determine their heights.
Data processing
- Processing the OSIF cloud satellite data of the buildings in PostGIS to get building heights and shadow generated at a time.
- Integrating the shadows produced by each building in QGIS.
AI Model Development
- We engineered an advanced AI model designed to optimize vehicle routing by analyzing key business parameters such as route path, route cost, completion time, and traffic conditions. Additionally, we integrated shadow pattern data to enhance route selection for maximized solar energy capture.
- To train our Artificial Intelligence model, we employed a Multiple Linear Regression technique. This approach allowed us to analyze and predict optimal vehicle routes by considering the previous variables.
- We successfully trained our model utilizing the multiple GPU available on Google Cloud Platform. This approach enabled us to leverage high-performance computing to significantly reduce training time and increase the efficiency of our model development process.
Backend Microservices Development
- We implemented backend microservices REST API with Python and Falcon to dynamically retrieve building shadow patterns from a PostGIS database in real time.
- Additional microservices were developed to interface with our AI model, facilitating the selection of optimal vehicle routes by incorporating shadow pattern analysis.
Frontend development and product integration
- We seamlessly integrated our MVP with Google Maps to display optimal vehicle routing options for users.
- Additionally, we developed a chatbot utilizing Google DialogFlow, designed to assist users in selecting among various route options.
What we achieved
We engineered a groundbreaking solution by leveraging satellite data to create an Artificial Intelligence model. This model precisely predicts the most efficient routes for solar-powered vehicles in real time, optimizing their energy utilization and enhancing operational efficiency.
Challenges, difficulties and solutions
The project faced numerous challenges and difficulties, overcome thanks to the enormous effort of the team.
Dataset and model training can account for over 60% of the total project duration
- Processing satellite data is complex and requires a very powerful computer, specially if you have not done it before. An 8TB high speed SSD is a must for processing, and some mechanical hard disks (20TB+) is needed to save all the satellite data.
- The processing time for getting the building heights of a small area of the city can take several days/weeks, and everything must be set up the right way for maximizing stability.
- The data analysis and preparation phase for creating an AI model is a critical yet time-consuming stage, often accounting for over 60% of the total project duration. This phase involves meticulous examination and structuring of the dataset to ensure it is optimally aligned with the objectives of the model. To enhance the model’s predictive efficiency, training should be conducted in incremental stages. This gradual approach allows for continuous refinement and optimization of the model’s performance.
Results
We successfully met our client’s objectives by developing an innovative solution that optimizes vehicle routing based on solar energy utilization, leading to significant cost reductions in vehicle routes. Our solution was seamlessly integrated into the solar vehicles, leveraging our sophisticated integration with Google Maps and enabling enhanced navigation capabilities.
Conclusion
The processing of satellite data, maps, and the development of Artificial Intelligence datasets and models necessitate expertise in their respective domains, constituting a significant portion of project time and requiring meticulous analysis. Through the sophisticated application of Artificial Intelligence algorithms, we have demonstrated the capacity to achieve remarkable outcomes. Our efforts culminated in the successful attainment of a complex goal, significantly enhancing our customer’s operational efficiency and reducing costs.
Even though something may seem complex or impossible, there is always a solution, and breaking down the problem helps in finding it.